DeepSeek-V3 is an open-weights model from DeepSeek, released December 26, 2024. It has an ECI score of 132 and ranks 154th of 240 tracked models. At the time of its release, it ranked 5th of 87 tracked models.
| Domain | Benchmark | ||||
|---|---|---|---|---|---|
ECI scoreBest score 173 | 132 | 121 | 141 | 133 | |
| Ranking | #154 | #182 | #117 | #149 | |
| Conceptual reasoning | SimpleBenchBest score 95% | ||||
| Long context | ForecastBenchBest score 66.7 | 62.4 | — | — | 62.9 |
| Mathematics | MATH Level 5Best score 98% | ||||
OTIS Mock AIME 2024-2025Best score 100% | |||||
| Science | ARC AI2Best score 95% | ||||
GPQA DiamondBest score 97% | |||||
| Software engineering | METR Time HorizonsBest score 17 hr 25 min | 18 min 28 sec | — | 38 min 50 sec | 20 min 31 sec |
Aider PolyglotBest score 88% | |||||
| Other | LiveBenchBest score 82% |